View Proposal

Proposer
Claudio Zito
Title
Robot Grasping
Goal
Develop ML algorithms for robot grasping
Description
This dissertation proposes the development of advanced contact-based algorithms tailored for improving the performance and reliability of robot grasping tasks. As robots increasingly integrate into diverse environments, the ability to accurately and efficiently grasp various objects becomes paramount. Current grasping methodologies often rely on visual or geometric information, which can be limiting in dynamic or unstructured settings. By focusing on the contact dynamics between the robotic gripper and objects, this research aims to enhance grasp robustness and adaptability, ensuring more effective operation in real-world applications. The project will be implemented and evaluated using the Robot Operating System 2 (ROS2) and can be conducted in simulation or on the Kinova Gen 3 Lite robot equipped with a 2-DOF gripper available in the lab. Introduction: The ability of robots to grasp objects is a critical aspect of autonomous functionality in fields such as manufacturing, healthcare, logistics, and service industries. Traditional grasping techniques, which rely heavily on visual perception or geometric shape analysis, may struggle to account for the complexities of real-world scenarios, such as object variability, surface friction, and unexpected environmental changes. This research seeks to address these challenges by developing contact-based algorithms that respond to real-time feedback from sensor data during the grasping process. Research Objectives: Algorithm Development: Design and implement contact-based algorithms using ROS2, leveraging sensory information (e.g., force, pressure, and tactile feedback) to enhance the adaptability and reliability of robot grasping. Modeling Contact Dynamics: Develop mathematical models to accurately simulate contact interactions between the 2-DOF gripper of the Kinova Gen 3 Lite and a variety of objects, considering factors like shape, material properties, and environmental conditions. Experimental Validation: Conduct extensive experimental studies that evaluate the performance of the developed algorithms, either in simulation or on the physical Kinova Gen 3 Lite robot, across various grasping scenarios, including handling rigid, soft, and deformable objects. Integration with Learning Techniques: Explore the integration of machine learning methods to optimize grasping strategies based on past experiences and predictive models. Methodology: The proposed research will involve a multi-disciplinary approach, which includes: Literature Review: Conducting a thorough review of existing grasping techniques and contact modeling approaches. Algorithm Design: Creating algorithms that utilize real-time sensor data to compute grasp strategies dynamically, specifically focusing on the capabilities of the ROS2 framework. Simulation and Testing: Utilizing robotic simulators to evaluate algorithm performance before real-world application. This phase will allow for parameter tuning and validation of the models in a simulated environment, followed by evaluation on the Kinova Gen 3 Lite to ensure practical applicability. Prototype Development: Implementing the algorithms on the Kinova Gen 3 Lite equipped with the 2-DOF gripper to perform grasping tasks and gather empirical data. Expected Contributions: Through the successful completion of this dissertation, it is anticipated that the project will contribute to: A set of innovative contact-based algorithms implemented in ROS2 that significantly improve robot grasping capabilities. New theoretical insights into contact dynamics between the Kinova Gen 3 Lite's gripper and various object types. Practical guidelines for implementing adaptable grasping strategies in real-world robotic systems. An extensive dataset of grasping performance across multiple conditions, which can serve as a reference for future research in the field. Conclusion: This dissertation will address a crucial gap in the existing robotic grasping technologies by focusing on contact-based methodologies implemented within the ROS2 framework. The outcomes are expected to enhance the functional capabilities of robots in diverse applications, paving the way for more intelligent and reliable robotic systems capable of operating in complex and unstructured environments. Furthermore, by integrating real-time feedback into the grasping process, this research will contribute to the broader field of robotics and automation, ultimately enhancing the interaction between robots and their environments. The evaluation, whether conducted in simulation or on the Kinova Gen 3 Lite robot, will ensure that findings are practically applicable and beneficial for advancing robot grasping technologies.
Resources
ROS2 Humble https://docs.ros.org/en/humble/index.html Kinova Gen3 Lite https://www.kinovarobotics.com/product/gen3-lite-robots https://github.com/Kinovarobotics/ros2_kortex Point Cloud Library https://github.com/ros-perception/perception_pcl
Background
Generative grasp synthesis https://arxiv.org/pdf/1906.11548
Url
Difficulty Level
High
Ethical Approval
None
Number Of Students
0
Supervisor
Claudio Zito
Keywords
robotics, robot grasping, machine learning
Degrees
Bachelor of Science in Computer Science
Master of Science in Artificial Intelligence
Master of Science in Human Robot Interaction
Master of Science in Robotics
Bachelor of Engineering in Robotics
Master of Science in Robotics with Industrial Application
Back to List